Your café menu and hours keep changing: how can the website and AI agree?

Which facts are stable and which need daily checks?

Put the address, directions and regular opening days on the public page. Give daily desserts, limited products and temporary closures an owner and last-checked date. An old article cannot prove that strawberry cake is in stock today. Without current inventory, say availability is unverified and point to today's announcement or staff. Google's special-hours guidance treats short exceptions separately from regular hours.

In the fictional example the café decides on Friday to close at four on Saturday. The operator updates the website, checks Google Business Profile special hours, refreshes the FAQ or website knowledge, then asks the public page and AI the same Saturday question. A social story alone leaves search visitors exposed to old information.

Classify each item as regular, exceptional or live. Regular facts receive a weekly owner review; exceptions are published with an effective date as soon as decided; live stock needs an on-duty updater. If the shop cannot maintain live stock, keep it out of automatic answers. This classification connects each answer to someone accountable for its currency.

Write a customer-facing example: 'Weekdays 10–6; Saturday closes at 4 for the event; Sunday follows regular hours. Check the counter for today's cakes.' This is more useful than a vague direction to the website, provided someone verified the dates. Remove an expired exception or mark it clearly as history; a stale special-hours FAQ becomes false knowledge.

How do page, FAQ and AI knowledge move together?

A kakinang microsite supports draft, preview and publication. Check the mobile view and enabled languages before release. FAQs can be edited directly; website knowledge synchronization changes the active version after a candidate completes; supported documents can be uploaded. A public page serves readers, while enabled knowledge serves AI retrieval. A draft edit or unfinished synchronization is not proof the assistant has learned the new menu.

Keep a change line with the old value, new value, effective time, operator, page check and AI check. After moving Sunday opening from nine to ten, ask both 'When do you open?' and 'Can I get coffee at nine on Sunday?' The second phrasing reveals a stale answer or a confusion between opening and last-order times. Google's people-first guidance stresses whether visitors can accomplish their purpose, not keyword repetition.

If the café also maintains a Google profile, social account and delivery listing, do not assume one edit updates every surface. Use a short per-channel checklist with the last review time. Publish an effective date on the primary page and assign a person to reconcile the rest. AI should report only verified website facts, not claim that every platform is synchronized.

Ask a colleague who did not edit the site to find today's hours and cakes from the home page. If the answer is buried three levels deep or the page and AI name the same item differently, repair navigation and naming. Product photos help recognition, but text still needs the name, supply date and relevant conditions. An uncaptioned menu photo is a weak source for search and accessible reading.

What should AI say while facts are still changing?

Approved regular menus are suitable for explanation. Sold-out items, limited batches and unpublished closures require current evidence. If the shop manages products and stock on its public catalog, visitors can inspect published items; a purchase still needs formal confirmation and rechecking. If today's menu lives only on a chalkboard, say the in-store notice is authoritative. Yesterday's photograph is not live stock.

Before opening, check today's hours and limited items. Weekly, inspect FAQs, the last successful synchronization and three common questions. Review holidays separately. When an answer is wrong, determine whether the page, synchronization, activation or source menu is stale. Correct the source first, then test direct and conversational wording. Prompt changes alone cannot repair an outdated menu.

Test exceptions: an unannounced seasonal drink should be described as unannounced; missing ingredient information should be referred to staff rather than guessed; temporary hours need an exact verified date. Keep expected and actual answers plus the person who corrected them. One accurate maintainable menu is more valuable than many similar pages that become stale.

Each month, sample ten menu and hours questions. Classify them as unpublished information, published but not retrieved, answered with an expired date, or really a request to reserve a table. Add a public answer for the first, inspect knowledge synchronization for the second, improve exception expiry for the third, and provide the right booking or staff route for the fourth. The classification identifies the broken step without blaming every issue on AI.

A frequent handoff failure occurs when on-duty staff know an item sold out but the page editor is away. Give the shift worker a simple way to mark today's item sold out and disable a purchasable option, or a named owner and a clear fallback answer if permissions prevent direct editing. Asking AI to say 'out of stock' while a product page still sells it creates a new contradiction.

Before a holiday, replay the visitor path on a phone: search the café, open its page, ask AI about tomorrow's hours and cake, then compare with the rota and preparation sheet. Suspend unsupported live promises until the records agree. This short review catches conflicts before a customer travels to the shop.

Separate same-day sellouts, week-long changes and permanent menu updates. For a sellout, change the sale state and ordering route first, then alert staff not to recommend it from an old FAQ. Special weekly hours need the site and external business listing updated. A permanent menu change belongs in the stable page and site knowledge. At each shift handover, compare the current item list with approved notices and dates. If asked whether a drink returns tomorrow, today's sellout note is insufficient; AI must not turn a temporary shortage into permanent discontinuation.